Survey Board / Swamp Map

A visual route through the portfolio.

Every map marker represents a key area of my work.

SWAMP LAB / ACTIVE QUEST

Nicholas Greiner

Computer Science @ University of Florida

C++ • Machine Learning Systems • Software Engineering

View Projects View Research Contact Me

CURRENT FOCUS

Building systems from fundamentals to agents

  • Classical ML in C++
  • VLA / robotics research
  • PvZRL generalist training

Base Camp / Character Profile

C++, machine-learning systems, and software engineering.

I am a UF computer science student, C++ Teaching Assistant, and Undergraduate AI Research Assistant.

My work spans numerical programming, reinforcement-learning infrastructure, and practical research pipelines. I have studied attribution-map signals for out-of-distribution detection and now work with vision-language-action models, OpenVLA, pi0.5, LIBERO, and failure analysis in robotics evaluation.

Outside of technical work, I enjoy spending time with my girlfriend and cats, meeting new people, and staying connected with long-time friends.

Analysis Dock / Featured Research

Attribution Maps for OOD Detection

Attribution-map workflow diagram with image inputs, heatmap output, and evaluation panel.

RESEARCH QUESTION

Researching whether neural network attribution maps can help distinguish in-distribution images from out-of-distribution samples using ResNet models, Integrated Gradients, and binary evaluation pipelines.

1 Generate attribution maps

Use Captum Integrated Gradients with ResNet-style image classification workflows.

2 Build binary datasets

Compare ID and OOD attribution-map samples through repeatable binary classification experiments.

3 Evaluate signal quality

Analyze AUROC, AUPRC, F1, and confusion-matrix behavior to guide iteration.

Workshop Outpost / Quest Log

Selected systems, numerical work, and shipped experiments.

A focused set of current projects, ordered by technical depth and backed by public repositories or working demos.

Decorative pixel-art strategy console representing the PvZRL state and action pipeline.
System illustration
Public / Active

PvZRL

A Windows reinforcement-learning and automation system that trains a single Adventure policy against PlantsVsZombiesRH. Gymnasium, MaskablePPO, behavior cloning, and a localhost C# bridge share a fixed, dynamically masked game contract.

  • 20k+Python + C# lines
  • 841 / 4,364actions / features
  • 500k+environment steps
  • Python
  • C#
  • MaskablePPO
  • Gymnasium
  • Behavior cloning
View on GitHub
Decorative pixel-art numerical-computing workbench with a matrix, regression plot, and modular instruments.
Pixel-art illustration
Active Library

Classical ML in C++

A C++20 and Eigen library for implementing machine-learning fundamentals from scratch. Linear regression supports both a normal-equation solution and batch gradient descent through reusable fit/predict interfaces, with modular regression metrics.

  • C++20
  • Eigen
  • Numerical methods
  • Linear regression
  • Model metrics
View on GitHub
NumberSense live application showing the first 24 times 19 mental-math lesson.
Live app capture
Live

NumberSense

A lightweight daily mental-math game about finding equivalent expressions and alternate solution strategies. A reducer drives progression while safe expression handling, curated teaching views, and versioned local persistence keep player math separate from authored guidance.

  • TypeScript
  • React
  • Vite
  • Reducer state
  • Safe expression parsing
View on GitHub Live Demo
Gator Gabber live application showing Spanish course controls and the conversation interface.
Live app capture
Live / Active

Gator Gabber

A full-stack Spanish conversation-practice app with text, speech, translation, syllabification, and course-aware context. Its React/FastAPI pipeline implements retrieval-augmented generation for SPN1130 text conversations; additional course-specific retrieval remains planned.

  • React
  • FastAPI
  • Python
  • RAG
  • Speech-to-text / TTS
View on GitHub Live Demo

SECONDARY BUILDS / SIDE QUESTS

Smaller builds with distinct roles.

Hardware Build

Raspberry Pi Sesame Robot

A Raspberry Pi adaptation of the Sesame quadruped using a PCA9685 controller and eight MG90S servos.

  • Raspberry Pi
  • PCA9685
  • Servo control
View on GitHub
Playable

Scrapstorm Overdrive

A browser-playable Pygame CE arena shooter with tank movement, waves, bosses, run upgrades, weapon evolutions, and persistent meta progression.

  • Python
  • Pygame CE
  • Pygbag
  • Game systems
Play Game

Lantern Route / Experience

Experience shaped by research, teaching, and practical builds.

A concise view of the applied work behind my research and software portfolio.

  1. Research desk with heatmap monitors and evaluation charts. 01 / Machine Learning Research

    Undergraduate AI Research Assistant

    Build PyTorch research pipelines for ResNet and Integrated Gradients ID/OOD analysis, then extend failure-analysis workflows to OpenVLA, pi0.5, and LIBERO.

  2. Mentoring table with coding notebook, route cards, and guidance markers. 02 / C++ Instruction

    C++ Teaching Assistant

    Lead discussion, office-hour, and code-review support around problem solving, debugging, object-oriented programming, and implementation fundamentals.

  3. Tutoring desk with notebook, graph board, calculator, and study cards. 03 / One-on-One Instruction

    Math Tutor

    Helped students develop confidence with math concepts through structured practice, clear explanations, and targeted feedback.

  4. Swamp workshop bench with laptop, project panels, and build tools. 04 / Applied Software Development

    Software Projects & Automation

    Build C++ numerical libraries, Python/C# reinforcement-learning infrastructure, web products, and game systems around clear contracts and iterative diagnostics.

Supply Depot / Skills Inventory

Tools grouped by how they are used.

Field Kit 01

Machine Learning

Model development, reinforcement learning, explanation methods, and evaluation.

  • PyTorch
  • ResNet
  • Integrated Gradients
  • MaskablePPO
  • Gymnasium
  • Behavior cloning
  • OOD detection
Field Kit 02

Programming

Product-focused implementation across interfaces, numerical code, and systems.

  • C++
  • Python
  • C#
  • TypeScript
  • React
  • FastAPI
  • HTML/CSS
Field Kit 03

Data / Tools

Numerical computing, developer tooling, and structured investigation.

  • Eigen
  • NumPy
  • pandas
  • TensorBoard
  • Git
  • GitHub
  • Visual Studio
  • Structured diagnostics
Field Kit 04

Research

Question framing and disciplined iteration for interpretable evidence.

  • Experiment design
  • Metric analysis
  • OpenVLA
  • pi0.5
  • LIBERO
  • Failure analysis

Final Checkpoint / Contact

Open to software engineering and AI research work.

Reach out directly, review public work on GitHub, or download my current résumé.

Swamp checkpoint dock with lantern and research outpost.